OpenAI의 다가오는 딸기 모델과 AI 산업 개발

### OpenAI의 스트로베리 모델과 애플의 최신 혁신 기술
* 🍓 신뢰할 수 있는 정보 유출자 지미 애플스에 따르면 **OpenAI는 향후 2주 내에 추론 중심의 AI 모델인 스트로베리를 ChatGPT의 일부로 출시**할 예정이라고 합니다.
* 🧠 스트로베리는 **추론 시간이 더 오래 걸리지만 고유한 사고, 추론, 계획 능력**을 통해 복잡한 쿼리를 처리하도록 설계된 ChatGPT 내의 독립형 제품이 될 것으로 예상됩니다.
* 💰 Strawberry의 가격은 아직 불확실하지만, **고급 기능을 반영하여 기존 OpenAI 챗봇 티어보다 높을 것**으로 예상됩니다.
* 📱 최근 Apple의 컨퍼런스에서 **새로운 iPhone 16 Pro Max에는 Apple Intelligence가 기본적으로 제공되지 않**는다는 사실이 밝혀져 많은 잠재적 구매자들을 실망시켰습니다.
* 🔍 Apple은 AI 모델의 온디바이스 프로세싱을 강화하여 필요한 경우에만 작업을 **ChatGPT로 오프로드**하는 것을 목표로 하고 있으며, 이는 더 뛰어난 Siri를 위한 비전에 부합합니다.
* ⚖ 한편, 엔비디아는 반독점 조사 중 법무부 소환에 직면해 있으며, **AI 칩 시장에서의 지배력에 필수적인 독점 CUDA 라이브러리**에 대한 우려가 부각되고 있습니다.
* 🚀 다른 소식으로는 **UAI가 엔터프라이즈 AI를 발전시키기 위해 5천만 달러를 모금**했으며, **샘 알트먼은 미국 내 AI 인프라에 대규모 투자**를 계획하고 있어 이 분야의 강력한 미래 성장을 예고하고 있습니다.
* 📈 마지막으로, **Find 405b의 출시와 딥 시크의 발전**은 저렴한 비용으로 뛰어난 성능을 보여주는 **딥 시크의 오픈 소스 모델**을 통해 AI 모델의 빠른 진화를 보여줍니다.

OpenAI의 Strawberry 모델 출시가 임박함에 따라 ChatGPT의 추론 기능이 향상될 것으로 기대됩니다.

open AI strawberry model is imminent. That's our first story for today. According to Reuters, we have open AI plans to release strawberry for ChatGPT in 2 weeks. And Jimmy Apples, the only reliable leaker, has mentioned it as well. Jimmy Apples last week: All Quiet on the Western Front, a heavy fog of skizo energy. I'm ready to be heard again. Let's see leaning forward then. Jimmy Apples 15 hours ago: This week I take a small step out of the cave of patience. I hope he is not forced to get back in that cave because I want to see some new stuff out of open AI. 3 hours ago: The age of patience is over. It's release season, so seems like it's coming soon. Let's take a look at the article. Open AI plans to release strawberry, its reasoning-focused artificial intelligence, as part of its ChatGPT service in the next 2 weeks, the information reported on Tuesday. And here's the article from the information. We should explain that while strawberry is part of ChatGPT, it's a standalone

offering exactly how it will be offered is unclear. One option is for strawberry to be included in the drop-down menu of AI models customers can pick from to power chat gbt, and it's quite different from the regular services, with some advantages and shortcomings. So, what differentiates strawberry is its ability to think, to reason, to plan, to take its time, and that is the key differentiator. It will actually take more inference time when you submit a prompt. So, the initial version will only be able to take in and produce text, not images, so not yet multimodal. Then there's pricing. Strawberry is likely to be priced differently from OpenAI's chatbot, which has free and subscription pricing tiers. They're not sure how it's going to be priced yet, but I think they're probably going to charge more for it. Strawberry is expected to be easier to use than GPT-4 for complex or multi-step queries. Now, apparently, some people have already used it. Some people who've used a strawberry prototype have complained that its

slightly better responses compared to OpenAI's currently released GPT-4.0 aren't worth the extra 10 to 20 seconds of waiting. And here's the thing: 95% of prompts and use cases can be accomplished with GPT-4.0 Mini, not even GPT-4.0. Then you layer on GPT-4.0 for the last 3 to 4%, and then finally maybe you need Strawberry for the most complex reasoning. Now, a week or so ago, The Information had reported that Strawberry is mainly used to generate synthetic data for Orion, OpenAI's next-generation frontier model. So, this is a little different than we thought was happening, but either way, we're getting something new from OpenAI soon, according to The Information, and of course, according to Jimmy Apples. And the next big story of this week is Apple's major announcements from yesterday. Apple had its conference yesterday. They made a number of announcements around the iPhone, the watch, and Apple intelligence. But here's the thing: even if you buy the brand new

iPhone 16 Pro Max. Whatever. Spend $1,200, $1,300 on this brand new hardware device, which really is just an incremental improvement over the 15, the 14, and so on. You still won't get Apple intelligence out of the box. It's not going to be shipped natively, and that is very disappointing. Now, you already know that I'm going to get the new phone. I'm going to install the beta, and I'm going to play around with it, and I'll review it for you, of course. And just a quick reminder of Apple intelligence. It is its own native model running on device for the most part. Now, whenever there are World Knowledge Questions where the local model just can't do it, it's going to offload it to Chat GPT in the cloud, but it will ask you first. This is the partnership with OpenAI. Now, I really like Apple's approach here. Do as much as you can on device and only offload it if you absolutely need to. And Apple actually has a larger version of their

model in their private Cloud before they even ask you to offload to open AI's models. I cannot wait to have a seriously capable Siri, an agent, an AI that is performing tasks, completing things on my behalf 24 hours a day. That is my perfect vision for the future of how I want to interact with artificial intelligence, and I really believe Apple and Google are both well-positioned to accomplish that because they both have the hardware in your pocket to actually interface with you. And the iPhone 16's camera will actually be able to do a lot of what the meta AI glasses can do. You take a picture of something, and it will tell you what you're looking at. So, in the information article, it says the iPhone 16 camera will also include visual intelligence, allowing customers to look up information about the scene in front of them, such as finding reviews for a restaurant they're standing in front of. I don't really use it all that often, I have to be honest. I love my meta AI glasses, but I don't use the visual AI.

all that much, and I actually don't use the voice AI all that much either. I use it all the time for taking pictures, though, but maybe this is going to change when we have it all built into our phones. We will see next, according to Bloomberg, Nvidia gets DOJ (Department of Justice) subpoena in escalating antitrust probe. Nvidia is the absolute monster in the AI world. They are one of the top three most valuable companies in the entire world. They basically sell the majority of all chips powering both fine-tuning, training, and inference, and they have this little library called CUDA, which is proprietary to Nvidia and is the software that essentially powers the GPUs. That is probably what they're getting in trouble for because the CUDA library is so powerful and so deeply ingrained in the Nvidia ecosystem that it's hard for anybody to justify buying other companies' chips simply because of the CUDA library. Now, even though Bloomberg reported this subpoena, it seems Nvidia is denying they ever got it, and it already affected

their stock price. So, it's interesting to see that a major publication like Bloomberg has reported it, but Nvidia denied it. So, of course, we're getting conflicting information right now. Next, we have a new state-of-the-art model on the S agent leaderboard. That is a coding AI leaderboard honeycomb. It's called Just Beat out Amazon Q with a 22.06% Devin score, which absolutely took the Internet by surprise. Got a 13.86% and honeycomb number one. So, here are the two honeycomb Founders, 19-year-old dropouts from MIT. Of course, they are in Y Combinator, and it is Jared Freeman, a YC partner, who is reporting this information. So, honeycomb isn't just the model; it's a full product that you can use right now as an AI programmer. Now, of course, trying to look at this through a bit more skepticism, Jared Freeman is obviously promoting this because he has a financial interest in doing so. So, here is Honeycomb. It does look extremely

similar to Devon. I don't believe it's open source, so that's a shame, but still very cool. Next, UAI raises $50 million and predicts more AI agents than people by 2025. That is less than a few months away. U.com, the AI-powered search and productivity platform, announced today it has raised $50 million in Series B funding to continue their work in the Enterprise AI world. It was led by Georgian, with participation from tech giants Nvidia and Salesforce Ventures, in addition to Day One Ventures. They've raised about $100 million so far. Now, I haven't used U.com, but going to their homepage, it looks exactly like ChatGPT, but I'm sure they have a bunch more functionality under the hood. Next, according to Bloomberg, Sam Altman plans on spending tens of billions of dollars building out AI infrastructure in the United States, which I love to hear. Now, in the Yahoo Finance article about the same topic, it says OpenAI Chief Executive Officer Sam Altman plans for a massive buildout of the machinery and systems needed for

artificial intelligence, and he is beginning with an effort in the United States slated to cost tens of billions of dollars, according to a person familiar with the matter. Alman had spent the early part of the year seeking the US government's blessing for the project, which aims to form a consortium of global investors to fund the costly physical infrastructure needed to support rapid AI development. Now we already know we don't have enough chips, and we also don't have enough energy to reach AGI, so the fact that we're getting this massive investment into the manufacturing process, into the chip-making process, is phenomenal. I love hearing it. Backers are likely to include investors in Canada, Korea, Japan, the UAE, and OpenAI also envisions other private companies being involved in the project. So of course, this is great news for the US, and I hope we continue to invest in building everything we need to reach AGI on our soil. Next, Grok is making lots of progress. They now have a vision model which is powered by their Grok chips.

which means it is incredibly fast. So, you can now try it in the API console. It is the Llama 1.57B model. So, simply upload an image, and you can ask any question about it, and it is lightning fast. And as a reminder, I'm an investor in Grock. Little disclosure are there and more news from Grock. They have just increased the speed at which you can run inference on the Llama models. On the Llama 3.1 70B model, they are reaching 544 tokens per second. That is astounding. Now, if you've watched this channel at all, you know I love AI agents, and inference speed is especially valuable when you're building AI agents because eventually, those agents can go off and accomplish tasks on your behalf 24 hours a day, and that inference speed is just going to allow them to accomplish so much more and more quickly. All right, next. This is a bit late, but Repet has announced Repet agents, which apparently everybody's using now instead of Cursor. Repet is a fantastic completely online

code editor and now they have AI built in natively, just like cursor, except it can build you anything in the cloud and deploy it easily. So, here's a little example of how it works. He put in a prompt, very short over here, submitted it. There's a plan of what to build. The code gets built in real time. It is so fast. All the files are being built out, and all of this can be deployed and hosted through Replit so easily. I've actually only used Replit a few times, but every time I use it, it's fantastic. So, look at that. With just a single two-sentence prompt, we now have a fully functional map application. So, really, really impressive. So, if you haven't checked it out, definitely check it out. I'll drop links to this in the description below. Next, according to the information, OpenAI has been considering higher price tiers for their subscriptions, all the way up to $2,000 per month. Now, if they achieve AGI, I'm willing to pay basically anything for it, but let's take a step

back from AGI for $2,000 a month. What would you be willing to pay for if AI could take 20% of my workload off my plate? I would absolutely pay $22,000 a month. Now, it would have to be consistent and reliable and high quality, which AI models are not quite there yet, but maybe soon. And so, executives have discussed high-price subscriptions for upcoming large language models such as Strawberry and a new flagship LLM dubbed Orion. We've already talked about both of those things extensively, so definitely check out those videos if you haven't seen them. The pricing discussions come as ChatGPT maker looks to raise billions of dollars in capital to make up for the billions of dollars it's losing per year. Now, anybody who thinks OpenAI is going out of business because they're losing so much money, I think you need to take another look at how revolutionary their technology is and really how a lot of tech companies work. They're going to lose money for a long time until they make money, and then they basically print

money. At least that's the hope. So let me know how much you'd be willing to pay for strawberry for Orion and then for AGI. Drop that in the comments. Next, a new model is launched, find 405b. This actually came out last week. I just haven't had a chance to talk about it. Find 405b obviously is a fine-tune of llama 3.1. 405b. It scores a 92% on human eval, matching Cloud 3.5 Sonet. Now, given the controversy with reflection 70b, I'm going to continue to look at these releases with a bit of skepticism. However, find has been releasing models for a while now, and they've actually released really good models in the past, and they all seem to be legit. We're introducing a new flagship model, find 405b, along with a new find instant model that offers Lightning Fast search speeds for all of your programming and curiosity questions. We're excited to announce the launch of find 405b, our new flagship model based on the excellent

metal Lama 3.1 405b. We've trained it to be a state-of-the-art model for programming and technical tasks. It can ingest 128k tokens of context with a 32k context window available at launch, and it is now available for paid pro users. Next, it seems OpenAI's cultural changes internally are actually sticking. In a post by Jason Quan, the CSO of OpenAI, weighing in on California Bill 10:47, here's what we said to colleagues last Friday. So, hi everyone, we've heard that there is a petition circulating among employees of AI Labs where people have the option to express support for CA 1047 Bill. We want you to know that if you want to sign the petition, you should feel free to. Part of what makes OpenAI work is that we have a diversity of personal views regardless of our company position. Okay, so great. So, they are finally allowing OpenAI employees to speak out and voice their own opinions, which is fine, which is good.

Do you know about California 1047? I've looked into it a little bit. Let me know what you think in the comments. Next, Samba Nova systems releases the world's fastest AI platform. Sova Cloud runs Llama 3.1 405b at 132 tokens per second at full precision. That is lightning fast. The Llama 3.1 70b at 570 tokens per second is comparable to Gro, but unfortunately right now Grock doesn't support 405b. They did briefly; it got overloaded because of popularity, and then they had to take it down. But I'm sure they're going to release it again soon. So, Samba Nova 10x faster inference than GPUs. So, you can definitely take a look at SNOA, use their cloud platform, and get lightning-fast speeds. Congratulations to them on their launch. And for our last story, the company Deep Seek has released a new model, and it is outstanding at coding and math. Here we can see a Deep Seek v2.5 open source and

As you could see, it beats basically every other model at reasoning, math, arithmetic, knowledge, coding. I should probably do a full test of it. Let me know if you want to see that in the comments below. And look at the cost per million tokens. It is 14 cents per million input tokens and 28 cents per million output tokens. That is so cheap. And if you want to host it yourself, it's open source. You can download it. So congratulations to Deep Seek for their release. So that's it for today. If you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

Strawberry Q* SOON, Apple Intelligence Updates, $2,000/mo ChatGPT, Replit Agents (AI News)

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Matthew Berman

OpenAI의 스트로베리 모델 출시 예정

OpenAI는 추론에 중점을 둔 AI 모델인 Strawberry를 향후 2주 내에 ChatGPT용으로 출시할 계획입니다.

스트로베리는 독립형 제품으로 제공될 예정이며, ChatGPT용 AI 모델의 드롭다운 메뉴에 포함될 가능성이 있습니다.

이 모델은 다른 가격 구조를 가지며 추론에 더 많은 시간이 소요될 것으로 예상되며, 복잡하거나 다단계 쿼리에 대해 더 나은 응답을 제공하는 것을 목표로 합니다.

애플의 아이폰 16 프로 맥스와 AI 통합

iPhone 16 Pro Max는 이전 모델에 비해 점진적으로 개선되었지만 출시 당시에는 기본 Apple 인텔리전스가 부족했습니다.

Apple의 AI 접근 방식은 기기에서 로컬 모델을 사용하며, 필요한 경우 클라우드의 Chat GPT로 작업을 오프로드하는 옵션을 포함합니다.

iPhone 16의 카메라는 시각적 인텔리전스를 제공하여 사용자가 주변 환경에 대한 정보를 찾을 수 있도록 도와주며, 엔비디아는 AI 칩 시장에서의 지배적인 위치로 인해 반독점 조사가 확대되는 가운데 법무부 소환에 직면해 있습니다.

AI 산업 업데이트: 주가, 신규 모델, 주요 투자 현황

엔비디아는 주가에 대한 블룸버그의 보도를 부인하여 상반된 정보를 야기했습니다.

코딩 AI 모델인 허니콤은 19세의 MIT 중퇴자들로 구성된 설립자들과 함께 아마존 Q를 제치고 S 에이전트 순위표에서 1위를 차지했습니다.

com은 AI 기반 플랫폼을 위해 5천만 달러를 모금했으며, 샘 알트먼은 빠른 AI 개발을 지원하기 위해 글로벌 투자자 컨소시엄을 목표로 미국 내 AI 인프라에 수백억 달러를 투자할 계획입니다.

그로크의 빠른 라마 모델과 리펫의 AI 에이전트

57B 모델은 초당 544개의 토큰을 추론하는 초고속 추론 기능을 제공하며, 현재 3.

1 70B 모델에서는 초당 544개의 토큰을 추론할 수 있습니다.

Repet은 원활한 온라인 코드 편집 및 배포 환경을 제공하는 AI 에이전트를 출시했습니다.

OpenAI는 혁신적인 기술을 위해 수십억 달러의 자금을 조달하는 것을 목표로 고가의 구독 티어를 고려하고 있습니다.

Find 405b 모델 출시와 OpenAI의 문화적 변화

사용자 평가에서 92%의 높은 점수를 받고 번개처럼 빠른 검색 속도를 제공하는 Llama 3.

1의 미세 조정 모델인 Find 405b가 출시되었습니다.

이 모델은 32,000개의 컨텍스트 창으로 128,000개의 토큰을 수집할 수 있으며 유료 프로 사용자만 사용할 수 있습니다.

OpenAI는 직원들이 캘리포니아 법안 10:47에 대한 의견과 지지를 표명할 수 있도록 하여 회사 내 다양한 개인적 견해를 장려하고 있습니다.

삼바 노바 시스템즈에서 초당 132개의 토큰을 처리하는 세계에서 가장 빠른 AI 플랫폼인 라마 3.

Grock의 405b 지원은 높은 수요로 인해 일시적으로 중단되었습니다.

5는 코딩과 수학에 탁월하며, 입력 토큰 100만 개당 14센트, 출력 토큰 28센트라는 인상적인 비용과 오픈 소스를 통해 자체 호스팅할 수 있는 옵션을 제공합니다.